Novel Class Generalization

Novel class generalization focuses on enabling machine learning models, particularly in few-shot learning scenarios, to accurately classify data from classes unseen during training. Current research emphasizes improving feature extraction through techniques like contrastive learning and knowledge distillation, alongside developing more adaptable distance metrics and compositional representation learning to better handle novel combinations of features. This area is crucial for advancing robust AI systems, particularly in applications like autonomous driving and fine-grained image recognition where encountering novel instances is common. Improved generalization capabilities are essential for building more reliable and adaptable AI models.

Papers

July 2, 2024

Unleash the Power of Local Representations for Few-Shot Classification
Shi Tang, Guiming Luo, Xinchen Ye, Zhiyi Xia
Classification Code Real Power Shot Classification Local Feature Local Representation Class Diversity Novel Class Generalization

May 26, 2024

When does compositional structure yield compositional generalization? A kernel theory
Samuel Lippl, Kim Stachenfeld
Representation Learning Compositional Generalization Compositional Structure Nonlinear Representation Associative Thinking Isolation Distributional Kernel Novel Class Generalization

June 29, 2023

Understanding the Overfitting of the Episodic Meta-training
Siqi Hui, Sanping Zhou, Ye deng, Jinjun Wang
Meta Learning Model Overfitting Meta Training Shot Classifier Novel Class Generalization

June 11, 2023

Compositional Prototypical Networks for Few-Shot Classification
Qiang Lyu, Weiqiang Wang
Classification Code Prototypical Network Feature Transfer Prototype Selection Novel Class Generalization

April 2, 2023

Learning with Fantasy: Semantic-Aware Virtual Contrastive Constraint for Few-Shot Class-Incremental Learning
Zeyin Song, Yifan Zhao, Yujun Shi, Peixi Peng, Li Yuan, Yonghong Tian
LeArning Abstract Contrastive Learning Class Incremental Learning Unseen Class Contrastive Constraint Novel Class Generalization

January 3, 2022

Novelty-based Generalization Evaluation for Traffic Light Detection
Arvind Kumar Shekar, Laureen Lake, Liang Gou, Liu Ren
Convolutional Neural Network Strong Generalization Generalization Performance Traffic Light Detection Novel Class Generalization